Mainframes still relevant to data centre

There are ten strong reasons why mainframe architecture is ideal for service-oriented data center environments. Mainframes offer several cost advantages as well as simplified manageability, data handling, security and more. In this tip, I will discuss the reasons for choosing the mainframe for your next-generation data center.

Internal network backplane/in-memory processing. When deploying a service-oriented architecture (SOA), one of biggest concerns for systems architects is message traffic overhead, followed by the transfer of large files. IBM's System z mainframe has a massive internal systems bus capable of supporting large volumes of high-speed network message traffic and transferring large volumes of data between service-oriented applications.

Also, note that the new IBM System z10 can transfer data between processor books (banks of processors contained in a book module) at a rate of 6 GB per second! This is 2.2 times faster than System z9 and 15 times faster than data transfer in a typical distributed computing environment. Knock off the overhead associated with issuing remote procedure calls (because procedure calls are issued locally) and the overhead associated with encryption (because local sharing doesn't require network-level encryption), and messaging and data transfer is even faster. There are huge performance and security advantages in handling program-to-program communications and data transfers inside a highly secure, centralised computing environment.

SOA infrastructure integration. IBM SOA infrastructure -- Web Services, XML, automated provisioning, development tools, etc. -- is closely integrated with System z operating environments and optimized for performance on System z architecture. These tools and software products can be found in IBM's Rational, Tivoli, WebSphere and DB2 product lines. Contrast this with SOA infrastructure from other vendors. If you buy from Hewlett-Packard or Sun, you'll most likely buy a mix of SOA infrastructure and development environments from companies such as Oracle, BEA and Symantec (Veritas). There are costs avoidance advantages in buying an integrated stack from a single vendor, particularly related to systems, application and database integration costs.

Advanced virtualisation with IBM z/VM. Virtualisation is the ability to pool and exploit unused physical/logical computing resources. IBM's z/VM operating environment is recognized by users, competitors and analysts as the gold standard for virtualisation. With z/VM and the z10 shared-everything architecture, IT managers and administrators can manage the virtualisation of processors, memory and I/O better than any other commercial system on the market today.

It should also be noted that IBM's z/VM has the ability to run very reliably at system utilisation rates of 90% or higher. Few other virtualisation offerings are this powerful or reliable. Capacity can be adjusted by turning additional processors on or off. Also, hardware and memory can be easily shared, making it possible to run multiple images that can share resources without getting in each other's way.

Energy consumption characteristics. Start by considering this: In a distributed system environment, every system has its own power supply (often redundant power supplies required for failover). Power supplies are notoriously inefficient, usually losing 25% of the power they receive when converting between AC and DC. Multiply all the power supplies by all the underprovisioned servers in a given distributed computing environment, and most IT managers will find that distributed systems do less work than what can be done on a mainframe but constantly waste 25% of the energy they receive. System z power supplies are more efficient than most distributed system power supplies, and because IT managers don't underprovision System z, much more work is done while using less power.

Next, add in savings attained through consolidation and virtualisation. Many IT departments have cut their server energy bills in half by consolidating and virtualising server workloads. Finally, factor in costs related to network interface cards, switches, routers and hubs. Kiss those devices goodbye and tally these savings when comparing System z and distributed system energy consumption. The composite picture is that mainframes are more energy efficient than distributed systems architectures for several reasons:

• They don't have to push large amounts of traffic over external networks;
• They have greater utilisation rates (meaning they don't need to burn as much power as distributed, underutilized server farms);
• They don't need myriad external network devices to support traffic flow.

Manageability. As enterprises shift from tightly coupled applications to loosely coupled SOAs, service interactions -- as well as underlying systems, storage and networking functions -- will need to be monitored and controlled. IBM offers a bevy of products that are well-suited to manage service-process flows over underlying, automatically provisioned virtualised systems infrastructures, including sophisticated business process management; intelligent orchestration (and associated libraries); automated policy-based provisioning; workflow management; service-level management; energy monitoring and management; and more. Many of these are part of IBM's Tivoli product offerings.

Real-time workload handling responsiveness. System z has been architected as a scalable, non-disruptive environment that enables transparent capacity increases to handle unexpected workloads. Distributed system environments usually handle this activity manually, driving-up management costs.

Security. Distributed computing environments often have thousands of access points, a variety of disparate network, computing, and storage systems, and no centralised point of control. Accordingly, IT departments spend a lot of their budgets and time locking down security access points, performing cross-system security integration, purchasing security software site licenses to cover hundreds or thousands of distributed nodes, and paying higher salaries and benefits for IT security personnel.

System z offers IT executives a chance to centralise IT security under IBM mainframe control. Integrated mainframe security can significantly reduce costs related to security software licensing and integration as well as security management personnel. It can also simplify compliance testing and eliminate the need to purchase expensive external public keys. IT executives can expect to save hundreds of thousands (if not millions) of dollars in hardware, software, testing and human resources-related costs by using IBM's mainframe security architecture.

Shared everything (and its ramifications for server availability). System z9 and System z10 share everything, including an enhanced I/O subsystem, main memory and virtualised resources. Important resources are not wasted as they are in distributed systems architectures. Note: Sharing these resources makes failover far simpler in mainframe environments than in distributed computing environments, where physical systems are usually assigned to pick up the pieces when a distributed tower computer fails.

Superior systems and application management. Mainframes have advanced systems and application management programs that simplify service management, operations, security and power management.

Real estate/floor space. Data centers are starting to run out of floor space. For IT buyers concerned about floor space, System z packs a lot of processing power into a relatively small footprint (as compared with the floor space that dozens of networked SMP or PC servers might occupy if equivalently configured).

Over the last decade or so, the application programming model has shifted from tightly coupled to loosely coupled. This loosely coupled model enables applications to readily interact with each other by requesting and fulfilling services. This approach, combined with industry program-to-program and data-sharing standards, has helped overcome interoperability issues and enabled a whole new generation of Web-based, highly flexible applications to come to fruition. To support this service-oriented application program model, changes are required in computer systems designs. The traditional model of overprovisioning towered servers needs to be thrown out the window and replaced with a model that emphasizes high resource utilisation and services management.

As for IBM's System z10 mainframe environment, I believe that Systems z10 represents the ideal model for service-oriented computing. System z10 can be highly optimized (with virtualisation to the extreme); IBM's SOA offerings have been tightly integrated with System z10, and System z10 has several technical advantages over other architectures that make it the best self-contained, centralised service-oriented data center in the computing world (bar none). As for parting words of wisdom, consider this: There are huge advantages in handling program-to-program communications and data transfers inside a highly secure, centralised computing environment. Systems architects who do not explore this alternative are doing their organisations a major disservice.